JP2002288675A - Method, program, and system for document identification - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a system capable of realizing a highly accurate identification, in the identification operation of a document by reducing the operation amount of an operator in the registration operation for the document of a dictionary. SOLUTION: This document identification system comprises an image input part 52 for storing the digital image of the document read through a scanner in an image memory 12, after correcting a skew of the document and removing blanks around the document; a characteristic extraction part 53 for detecting the size of the document from the digital image in the image memory 12, detecting the character patterns in unit of character string, converting the character patterns into a character amount in units of character string, and holding the character amount in a character memory 13; a common character extracting part 55 for extracting the character amount common to a plurality sets of character strings from the character amount of the character strings in the character memory 13 and registering the extracted character amount in a dictionary memory 15, when the format of the document to be inputted is registered; and a collation part 54 for collating the character amount of the character string of the document to be processed in the character memory 13 with the character amount of the character string for each document in the dictionary memory 15, when the format of the document to be inputted is identified.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for identifying a form, and more particularly to a system for identifying a form when a form is input.
The present invention relates to a form identification method, a form identification program, and a form identification method for formatting and registering forms having different contents, or for identifying the type of format of a registered form.

[0002]

2. Description of the Related Art An example of a conventional form identification method is disclosed in
No. 1-282959. This conventional form identification system includes an image input unit, an image storage unit,
Form position detecting means for detecting a form position in an image, coordinate converting means for converting the position of a character string registered in the dictionary storage unit into the coordinate system of the image storing unit, and characters obtained by the coordinate converting unit Character coordinate correcting means for correcting the difference between the position of the column and the position of the actual character string in the image in character units; and a figure matching means for matching the dictionary and the character pattern in the image to be processed in character units. It is composed of

A conventional form identification system having such a configuration operates as follows. That is, a digital image obtained by optically scanning a form, photoelectrically converting the form, and digitizing the form is stored in the image storage unit. For the digital image of the form, first, the position of the four sides of the form is detected by the form position detecting means, and the position of the form in the image is obtained. Next, in the coordinate conversion means, the position information in the digital image is obtained by using the position information of the form obtained by the form position detecting means in the form unit of the pre-print character string of the form held in the dictionary. The coordinates are transformed into information. Next, the character coordinate correcting means corrects the positional deviation between the coordinates of the character pattern registered in the dictionary converted by the coordinate converting means and the coordinates of the character pattern in the actual image. Further, in the graphic matching means, the feature amount of the character pattern registered in the dictionary is compared with the feature amount of the character pattern in the image. After all character patterns are collated for each form type registered in the dictionary, the form type having the highest similarity is determined as the form type of the form to be processed.

[0004]

However, this prior art has the following problems. A first problem is that in the case of a character string with a narrow character interval, the character pattern cutout performed in the character coordinate correcting means fails, and the form identification may not be performed in some cases. This problem is remarkable in the case of a character string having a small character size and a narrow character interval.

[0005] The second problem is that the operator has to specify a character string to be registered in a dictionary place, which imposes a heavy burden on the operator.

SUMMARY OF THE INVENTION In view of the above problems, an object of the present invention is to reduce the amount of work of an operator in registering a form in a dictionary.
An object of the present invention is to provide a system that realizes highly accurate identification in a form identification operation.

[0007]

In order to achieve the above object, a form identification method according to the present invention is a form identification method for identifying a form by an information processing apparatus. A character string is detected from the image, coordinate information for each detected character string is extracted, and a window area that limits a rectangular feature extraction range having the same height as the character string is displayed on the character string. The feature amount is extracted while moving along the line, and the coordinate information of the extracted character string and the feature amount are registered in a memory to create a form identification format.

A form identification method according to the present invention is a form identification method for identifying a form by an information processing apparatus, wherein a character string is respectively converted from a digital image of a plurality of completed forms in the same format and having different contents. Is detected, coordinate information for each detected character string is extracted, and a window area for limiting a rectangular feature extraction range having the same height as the character string is moved along the character string along the character string. While extracting the characteristic amount, and comparing and checking whether the coordinate information and the characteristic amount for each of the extracted character strings match in all of the plurality of filled-out forms having the same input format and different entry contents. It is characterized in that a form identification format is created by registering, in a memory, coordinate information and a feature amount of a character string having the same coordinate information and feature amount.

A form identification method according to the present invention is a form identification method for identifying a form by an information processing apparatus, wherein a character string is detected from a digital image of an input form, and coordinates of each detected character string are detected. The information and the characteristic amount are extracted, and the type of the form is identified by comparing the coordinate information and the characteristic amount of the extracted character string with the coordinate information and the characteristic amount of the character string registered in advance for each type of the form. It is characterized by:

The form identification method according to the present invention is a form identification method for identifying a form by an information processing apparatus. When a blank form is entered in a new format, a digital image of the entered form is converted to a second form. A first holding step of holding the data in the first memory, a size measuring step of measuring a form size from the digital image held in the first memory, and a character string from the digital image held in the first memory. Detecting and calculating coordinate information of the detected character string, and further moving a window area that limits a rectangular feature extraction range having the same height as the character string along the character string along the character string A feature value calculating step of calculating a feature value; a second holding step of storing the calculated coordinate information and the feature value in a second memory; A first repetition step of repeating the calculation step and the second holding step; and, when the first repetition step is completed, the form size held in the first memory and the total size held in the second memory. And registering a feature amount of each character string for each coordinate information in the third memory.

The form identification method according to the present invention is a form identification method for identifying a form to an information processing apparatus. When a form written in a format is input, a digital image of the input form is converted to a first image. A first storing step of storing the form in the memory, a generating step of generating an identification code for identifying the input form, and a size for measuring a form size from the digital image stored in the first memory. A measuring step, detecting a character string from the digital image held in the first memory, calculating coordinate information of the detected character string, and extracting a rectangular feature having the same height as the character string A feature amount calculating step of calculating a feature amount while moving the window region for limiting the range along the character string on the character string; and storing the calculated coordinate information and the feature amount in a second memory. Second holding on the basis of the identification code
And a first repetition step of repeating the feature amount calculation step and the second retention step until the presence of a character string disappears, and the same as the input form after the end of the first repetition step. When a form filled in in a format is input, a second repetition step of repeating the generation step, the size measurement step, the feature amount calculation step, the second holding step, and the first repetition step, After the end of the second repetition step, the coordinate information of the character string in the certain identification code stored in the second memory matches the coordinate information of the character string in another identification code stored in the second memory. And if there is a match in the comparison step, the similarity between the feature quantity corresponding to the matched coordinate information is determined. An extraction step of extracting all the coordinate information and the feature amount of the character string as a match if the value is equal to or more than a predetermined value, and extracting the feature amount for each coordinate information of all the extracted character strings after the extraction. Registering the form size written in the second memory and the third memory.

A form identification method according to the present invention is a form identification method for identifying a form by an information processing apparatus. When a form written in a format is input, a digital image of the input form is converted into a first digital image. A first storing step of storing the digital image stored in the first memory, a size measuring step of measuring a form size from the digital image stored in the first memory, and detecting a character string from the digital image stored in the first memory Then, the coordinate information of the detected character string is calculated, and further, the window area that limits the rectangular feature extraction range having the same height as the character string is moved along the character string along the character string. A feature amount calculating step of calculating the amount, a second holding step of holding the calculated coordinate information and the feature amount in a second memory, and the feature amount calculating step until the character string is no longer present. A first repetition step of repeating the step and the second holding step; a coordinate of a character string held in the second memory after the first repetition step is completed; and a character string stored in the second memory. A comparison step of comparing coordinate information of a character string of a form registered in the third memory with the same form size as that of the form, and a similarity between the characteristic amounts of the character strings if the character strings match in the comparison step Is calculated, and if the value is equal to or greater than a predetermined value, it is determined that the character strings match, and an identification step for identifying the type of the form is included.

A form identification program according to the present invention is a form identification program for identifying a form to be processed by a computer. When a blank form is entered in a new format, a digital image of the entered form is created. In a first memory, a size measuring step of measuring a form size from the digital image held in the first memory, and a character from the digital image held in the first memory. Detects a column, calculates the coordinate information of the detected character string, and moves a window area that limits the rectangular feature extraction range with the same height as the character string along the character string along the character string A feature amount calculating step of calculating a feature amount while performing the calculation, a second holding step of holding the calculated coordinate information and the feature amount in a second memory, A first repeating step of repeating said second holding step and the feature amount calculating step until the resident is eliminated, the first of said repeating step is completed and form size held in said first memory a second
And a registration step of registering, in a third memory, the feature amounts of all the character strings stored in the memory for each piece of coordinate information in the third memory.

A form identification program according to the present invention is a form identification program for identifying a form to be processed by a computer. When a form written in a format is input, a digital image of the input form is converted. A first holding step of holding in a first memory, a generating step of generating an identification code for identifying the input form, and measuring a form size from a digital image held in the first memory A size measuring step of detecting a character string from the digital image held in the first memory, calculating coordinate information of the detected character string, and further calculating a rectangular shape having the same height as the character string. A feature amount calculating step of calculating a feature amount while moving a window region for limiting the feature extraction range on the character string along the character string, and the calculated coordinates A second holding step of holding the distribution and the feature quantity on the basis of the identification code in a second memory,
A first repetition step of repeating the feature amount calculation step and the second holding step until there is no longer a character string; and, after completion of the first repetition step, data is entered in the same format as the input form. When a form is input, a second repetition step of repeating the generation step, the size measurement step, the feature amount calculation step, the second holding step, and the first repetition step, and the second repetition step After the step, the second
A comparison step of comparing all coordinate information of a character string in a certain identification code held in the memory with the coordinate information of a character string in another identification code held in the second memory; If there is a match in the comparison step, the similarity to the feature amount corresponding to the matched coordinate information is calculated, and if the value is equal to or greater than a predetermined value, the coordinate information of the character string and all the feature amounts are determined to be matched. A process including an extraction step of extracting, and a registration step of registering, in a third memory, a feature amount for each coordinate information of all extracted character strings and a form size written in the second memory after the extraction is completed. Is executed by the computer.

The form identification program according to the present invention is a form identification program for identifying a form to be processed by a computer, and when a form written in a format is input, a digital image of the input form is converted. A first holding step of holding in a first memory;
A size measuring step of measuring a form size from the digital image held in the memory; detecting a character string from the digital image held in the first memory; calculating coordinate information of the detected character string; A feature amount calculating step of calculating a feature amount while moving a window region for limiting a rectangular feature extraction range having the same height as the character string along the character string on the character string; and the calculated coordinate information. A second storing step of storing the characteristic amount and the characteristic amount in a second memory; a first repeating step of repeating the characteristic amount calculating step and the second storing step until there is no longer a character string; After the first repetition step, the coordinates of the character string held in the second memory and the form size stored in the second memory are registered in the third memory with the same form size as the form size stored in the second memory. A comparing step of comparing the coordinate information of the character string of the form with the document, and calculating the similarity between the characteristic amounts of the character strings when the character strings match in the comparing step, and if the character string is equal to or more than a predetermined value, the character string Is determined to match, and an identification step for identifying the type of the form is executed by the computer.

According to another aspect of the present invention, there is provided a form identification method for identifying a form, comprising: an input device for converting the form into a digital image when the form is input; and a control device for performing a form identification process. An image input means for storing a digital image of the form read from the input device, and detecting the size of the form from the digital image of the form held by the image input means and detecting the character pattern in units of character strings. A feature extraction unit that converts the character string into a feature amount and holds the same, and a common format for each set from a plurality of sets of character strings output from the feature extraction unit when the format of the input form is registered. A common feature extraction unit for extracting a feature amount and registering it in the dictionary memory; and a feature extraction unit for identifying the format of the input form. It is characterized by having a matching means for matching the feature quantity of the feature with a pre-character string held in each document in said dictionary memory string of the form to be processed to be output.

[0017]

Next, embodiments of the present invention will be described with reference to the drawings. Referring to FIG. 2, according to a first embodiment of the present invention, a control device 1, which is an information processing device (computer) such as a personal computer, and an image scanner 2 that optically scans a form surface to generate a digital image
A keyboard 3 for inputting characters and the like to the control device 1; and a display device 4 (for example,
CRT, LCD, etc.). The image scanner 2 may use an input device such as a digital camera that converts input image information into a digital image.

The control device 1 comprises a control unit 11 operated by a program (not shown) by a processor (CPU),
An image memory 12 for holding a digital image of a form, a feature memory 13 for holding a plurality of sets of character string feature amounts detected from forms of the same format having different entry contents by a feature extracting unit 53 (described later), And a dictionary memory 1 for storing the state of the control during operation of the control unit 11 including the mode at the time of the reading operation, and the feature amount of the character string together with the coordinates of the character string using the form type as an index.
5, a program memory 16 storing a form identification program, and an interface unit 17 for exchanging data (for example, controlling with a USB interface) between the image scanner 2, the keyboard 3, and the display device 4. The image memory 12, the feature memory 13,
And the control memory is a memory from which data can be read and written [the memory in this case is any type of internal memory (for example, RAM) and an external memory (for example, magnetic disk)]. It may be provided in each area. Also, the program memory 16
A memory that can read and write data,
It is desirable to use a non-volatile memory (for example, an EPROM, a magnetic disk, or the like) so as not to be lost even when the power is turned off.

The internal configuration of the control unit 11 shows a functional configuration when the control unit 11 executes the form identification program in the program memory 16, and controls the overall operation when the form identification program is executed. A common unit 51 for performing;
An image input unit 52 that performs form inclination correction and margin removal on a form digital image read from the image scanner 2 and then holds (stores) the image in the image memory 12.
A feature extraction unit that detects the size of a form from a digital image of the form stored in the image memory 12, detects a character pattern in units of character strings, converts the character pattern into character amounts in units of character strings, and stores the characteristic amount in the characteristic memory 13. 53 and feature memory 1
A common feature extracting unit 55 that selects only the feature amount of a character string common to each form from the feature amounts of the character strings detected from a plurality of forms having different entry contents in the same format and written in the dictionary memory 15, The character string feature amount of the form to be processed output from the feature extracting unit 53 is compared with the character string feature amount held for each form type in the dictionary memory 15, and the processing target output from the feature extracting unit 53 is compared. A collating unit 54 for determining and outputting a form type having the highest similarity to the character string characteristics of the form
And

A feature extraction unit 53 detects a form size from a digital image of the form stored in the image memory 12, and a character pattern from a digital image of the form stored in the image memory 12. And a character string feature detection unit 532 for converting the character string into a feature amount in a character string unit and writing the feature amount in the feature amount memory 13.

Referring to FIG. 1, when the control unit 11 executes the form program stored in the program memory 16 of the control device 1 of FIG. 1, the function blocks in the control unit 11 and the memories 12 to 15 It is the schematic which showed the relationship of.

Referring to FIG. 2, a first embodiment of the present invention will be described. Under the common unit 51, the image input unit 52
After reading the digital image obtained by sampling the surface of the form obtained by the image scanner 2 in a grid pattern and quantizing the density value, the inclination of the form in the digital image is corrected and margins are removed. Image to image memory 1
2 is stored. In this case, the process of correcting the inclination of the form and removing the margin around the form in the image input unit 52 is described with reference to FIG.
This will be described with reference to FIG.

FIG. 9 is an example of a digital image f obtained by the image scanner 2. It is assumed that the image is a binary image in which the pixel value of a peripheral portion of the form is 1, the pixel value of a portion where nothing is written on the form is 0, and the pixel value of a portion where a character is written on the form is 1. . However, a pixel having a value of 1 is a black pixel,
Pixel is also called a white pixel. The coordinate system of the image f is an xy coordinate system with the origin O at the upper left corner of the image, the + X axis in the right direction from the origin, and the + Y axis in the downward direction from the origin. Also, the inclination of the form in the image f is within a range of ± 45 degrees at the maximum (excluding +45 degrees and -45 degrees). First, the image input unit 52 detects the coordinate values of the four corners of the form in the image f. More specifically, for example, as shown in FIG. 9, the image f is scanned in an oblique direction of 45 degrees or 135 degrees from each of the four corners of the image f, and a black pixel representing the background represents a blank space on the document. The coordinates of the first point that changes to a white pixel are detected. Thus, the four corners of the form are detected by the image input unit 52. Here, the upper left corner of the form obtained as described above is P1 (x1, y1), the upper right corner is P2 (x2, y2), the lower left corner is P3 (x3, y3), and the lower right corner is P4 (x4, y4). y4). The inclination θ of the form in the image f is obtained as arctan ((x2-x1) / (y2-y1)) from the coordinate values of P1 and P2, for example.
The area on the form is based on equations representing the top, bottom, left, and right sides of the form, and the inequalities (1) (2) (3) (4): y ≧ ((y2-y1) / ( x2-x1)) x + (x1-y2-x2-y1) / (x1-x2) ... (1) y ≤ ((y4-y3) / (x4-x3)) x + (x3-y4 −x4 · y3) / (x3−x4) (2) x ≧ ((x3−x1) / (y3−y1)) · y + (y1 · x3−y3 · x1) / (y1−y3)・ ・ ・ ・ (3) x ≦ ((x4-x2) / (y4-y2)) ・ y + (y2 ・ x4-y4 ・ x2) / (y2-y4) ・ ・ ・ ・ Area that satisfies all of (4) Can be expressed as Also, form width W
And height H are respectively W = SQRT ((x2-x1) ・ (x2-x1) + (y2-y1) ・ (y2-y1)) ・ ・ ・ ・ (5) H = SQRT ((x3-x1 ) · (X3−x1) + (y3−y1) · (y3−y1)) (6) Here, SQRT (x) represents an operation for obtaining the square root of x. In the conversion of the skew correction and the margin removal, an equation for calculating the coordinate value (x, y) of the image f before the conversion from the coordinate value (X, Y) of the image g after the conversion is: cos θ · X−sin θ · Y + x1 = x · (7) It can be performed by using sin θ · X + cos θ · Y + y1 = y (8) That is, the values of X and Y are changed in the range of X = {X | 0 ≦ X ≦ W} and Y = {Y | 0 ≦ Y ≦ H}, respectively, and the converted image g (X, Y) is The pixel value is calculated by the equations (7) and (8).
By converting the image f (x, y) before conversion. Here, when the values of x and y do not become integers, values rounded off are used. The position on the image after the conversion is represented by a coordinate system with the origin at the upper left corner of the form, + X from the origin to the right and + Y from the origin to + Y. In the above description, even when the image obtained by the scanner is not a binary image but a grayscale image, in the conversion of the inclination correction and the margin removal, the image is once converted into a binary image by an appropriate threshold value and the coordinate values of the four corners of the form are calculated. After the detection, it can be applied by converting the original gray-scale image according to equations (7) and (8).

The image memory 12 holds an image obtained by subjecting the input digital image of the form to inclination correction and margin removal. After that, the image input unit 52 temporarily returns the control to the common unit 51 in order to transfer the control to the feature extraction unit 53.

The feature extracting unit 53 that has taken over the control from the common unit 51 detects the form size with the form size detecting unit 531, and the character string feature detecting unit 532 prints the form from the digital image of the form held in the image memory 12. A character pattern is detected for each character string, and is converted into a feature amount for each character string. It should be noted that there have been many reports on the detection of the character string unit of the print character pattern in this case. For example, first, for the binary image held in the image memory, a labeling process is performed on a connected component unit of a black pixel (a label value is assigned to a connected component unit of a black pixel to distinguish a connected component of each black pixel. Processing to make it possible to handle, and close in the vertical and horizontal directions in the obtained label graphic component (existing at a distance of about several pixels)
Are combined into a new label graphic component.
Next, the width and height of each of the label graphic components that have undergone the combining process are measured to select a label graphic component having a character-like size. At this time, regarding the width, a label graphic component having a value corresponding to the width of several characters is also selected as a character in consideration of the fact that characters with a narrow character interval are connected in the combining process. With respect to the label graphic components selected in this way, those having a shorter distance in the x direction and a smaller y coordinate value at the upper end and the lower end are sequentially grouped. When there are no more label graphic components that can be grouped, the grouping process ends, and each set of grouped label graphic components is used as a character string pattern.

The character string pattern detecting unit 532 of the characteristic extracting unit 53 applies the character string pattern detected as described above.
Extracts a feature amount in character string units. As shown in FIG. 7, the character string unit has a width of about 2 pixels to about 20 pixels (in the case of an image having a resolution of about 200 DPI) on the target character string pattern and has the same height as the character string. While moving the feature extraction window area having the following in the character string direction (horizontal direction in the horizontal writing character string) in units of one pixel to several tens of pixels, the feature of the character pattern included in the feature extraction window area is moved. Is used. At this time, as shown in FIG.
It is divided into (about 9) (5 in FIG. 10) partial areas.
The features of the character pattern are totaled for each of the partial areas.

The feature of the character pattern is a tangent direction (or a direction orthogonal to the tangent direction) to a pixel at a boundary between the character pattern and its background. The tangent direction is quantized in the N (for example, 8) direction, and the number of pixels is counted in each of the quantized directions for each partial region set in the window region as shown in FIG. Furthermore, normalization is performed for each direction in each partial region so that the total number of boundary pixels detected in one window region is K (for example, 100). That is,
The total number of boundary pixels actually detected in a certain window area is K
1. If the number of pixels counted in one direction in a certain partial area of the window area is M, the value of M is M
It is normalized as × K / K1.

As another characteristic of the character pattern, the density of black pixels in the partial area can be used. For example, "
Variations in the black pixel density of a character string in which Arabic numerals 1 are lined up, such as 111 ″, are quantified assuming that the divided area in the window area is 1 (the maximum ratio of the density of black pixels in the window area is 100). The pattern is “0, 10, 30, 90, 10
0,90,30,10,0,0,0,10,30,9
0,100,90,30,0,0,0,10,30,9
0, 100, 90, 30, 10 ".

If the mode at the time of the form reading operation when the mode is read from the control memory 14 is the form registration mode and the blank mode when the mode is read from the control memory 14, the feature extracting unit 53 determines the detected form size and the detected character string for each character string. The characteristic amount is stored in the characteristic amount memory 13, and the control is temporarily returned to the common unit 51 in order to transfer the control to the common characteristic extracting unit 55. In this case, the feature amount memory 13 holds the size of the form obtained by the feature extraction unit 53 and the feature amount in character string units. If the mode at the time of the form reading operation is the form registration mode and the filled-in mode, the feature extracting unit 53 repeats the form entered in the same format and having different contents in the same format until the instruction of the operator. A feature amount is extracted, and a form size and a set of feature amounts per character string detected for each form are stored in the feature memory 13. In this case, the feature amount memory 13 holds the size of the form obtained by the feature extraction unit 53 and a plurality of sets of character string unit feature amounts extracted from forms having different entries in the same format.
The feature memory 13 stores a plurality of feature amounts of a character string detected from a plurality of forms having the same format and different entry contents for each form (in this case, an unused identification code is automatically added to the dictionary memory 15). It has a capacity to hold. In addition,
Holding (or registering) a feature amount for each character string (or for each character string) in the feature amount memory 13 or the dictionary memory means that in the present invention, the coordinate information of the character string “for example, the coordinates of the upper left corner (for example, (Including X, Y coordinates), horizontal length, and vertical height ”corresponds to a character string, and holds (or registers) a feature amount (pattern) for each piece of coordinate information of the character string.
Means to do. In this case, storing (or registering) a feature amount for each coordinate means that the coordinate information is written in a memory that requires it, and the feature amount is written in comparison with the written coordinate information. .

When the mode at the time of the form reading operation when the mode is read from the control memory 14 is the form registration mode, the common unit 51 takes over the control to the common feature extracting unit 55. Common feature extraction unit 5 taking over control from common unit 51
Reference numeral 5 extracts the coordinates of the character string and the character string characteristics common to the forms having different entries in the same format stored in the characteristic amount memory 13 based on the character string coordinates and the character string characteristic amounts. If the difference between the coordinate values of the character strings is within a certain value (about several pixels), the coordinate values are considered to be the same, and the similarity between the character string features, that is, the feature vector in which the respective feature amounts are arranged in one column is u. , V, u · v / ‖u
If the value calculated by {v} (where * represents an inner product operation) is equal to or greater than a certain value (for example, 0.95), it is regarded as the same. However, it is also possible to determine the identity of the character string features by applying various conventional character recognition methods.

The extracted characteristic amount of the common character string is registered with the form identification code automatically added to the dictionary memory 15 as an index. In this case, the form size held in the feature memory 13 is also registered in the dictionary memory 15.
That is, the dictionary memory 15 holds the form size and the feature amount of each coordinate value of the detected character string (each character string unit) using the form identification code as an index.

If the mode is read out from the control memory 14 and the mode is the form identification mode, the feature extracting unit 53 returns the control to the common unit 51 once to pass the control to the collating unit 54.

Collation unit 5 taking over control from common unit 51
Reference numeral 4 denotes a form type and a coordinate value of a character string detected from a digital image of a form to be processed output from the feature extracting unit 53 in the form recognition mode, and a feature amount of the form, registered in the dictionary memory 15. And collate sequentially. It has the same size as the form to be processed, the coordinate information of all the character strings registered in the form type has the same coordinate value as any character string on the form to be processed, and If the feature amounts are the same, the form to be processed is determined to be of the form type, and the determination result is output. For the determination of the identity of the coordinate values of the character strings and the identity of the feature amounts of the character strings, the determination method described in the description of the common feature extraction unit is used.

Next, FIGS. 1 and 2 are replaced with FIGS.
The overall operation of the embodiment of the present invention will be described with reference to the flowchart of FIG.

The operation for reading a form according to the present invention is divided into a form registration mode and a form identification mode. The form registration mode is a mode for registering a feature amount for form identification in the dictionary memory, and is further divided into a blank mode and a filled mode. The blank mode is a process for registering the feature amount of the form in the dictionary memory using an unused form, and the completed mode registers the feature amount of the form in the dictionary memory using a plurality of completed forms. It is processing for. The form identification mode is a process in which a feature amount registered in the dictionary memory is compared with a feature amount detected from the form to be processed, and the type of the form to be processed is determined.

First, the operator starts up the control device 1,
When the form identification program is executed from the keyboard 3,
The control unit 11 of the control device 1 reads out and executes the form identification program stored in the program memory 16. Then, the common unit 51 of the control unit 11 displays a menu screen on the display device 4 via the interface unit 17. This menu screen includes items for prompting the user to select whether to perform processing in the form registration mode or in the form identification mode. Furthermore, there are two types of form registration modes: an unfilled mode (new form: only the form format is filled) and a filled mode (filled form: filled in according to the form format). It can be selected from the menu screen.

When the operator uses the keyboard 3 to select and execute “non-form mode in form registration mode” from the menu screen of the display unit 4, the common unit 5 of the control unit 11
In the mode area of the control memory 14, "No entry mode in the form registration mode (for example," 00 "in hexadecimal notation mode in the form registration mode," 0F "in fill mode in the form registration mode, The form identification mode is set to "FF" in hexadecimal notation). " After registration, common part 5
1 is waiting for a form to be input.

The operation of the form registration mode in the non-entry mode will be described below. The operator operates the image scanner 2
If you enter a blank form from the
Generates a digital image of a form by optically scanning the form surface and performing photoelectric conversion. The generated digital image of the form is read by the image input unit 52 via the common unit 51, and after correcting the form inclination and removing margins around the form, the digital image is stored in the image memory 12 (step A0 in FIG. 3).
1). After being stored in the image memory 12, the image input unit 52
Since the control is passed to the feature extraction unit 53, the common unit 51
Return control once.

The feature extraction unit 53, which has taken over the control from the control unit 51, randomly generates an identification code not registered in the dictionary memory in order to identify the read form.
The form size detection unit 531 measures the size of the form held in the image memory 12, and writes the generated identification code and the measured form size to the feature memory 13 (step A02). Further, the feature extraction unit 53 causes the character string feature detection unit 532 to detect the character pattern of the form held in the image memory 12 in units of character strings, and calculates a feature amount from the detected character pattern in units of character strings. (Steps A03 and A04). In this case, the feature extraction unit 5
The character string feature detection unit 532 for each character string for which the feature amount is calculated indicates coordinate information [eg, upper left corner coordinates (X, Y coordinates from the upper left corner) and horizontal length and height: character string Indicating the position of the area of the area] is calculated. This coordinate information is used for identifying a character string. The feature extracting unit 53 writes the calculated coordinate information and the calculated feature amount into the feature amount memory 13 as an index using the identification code generated earlier (step A05). In this way, the feature extracting unit 53 determines in step A0 until there is no more character string.
The steps 3 to A05 are repeated (step A06). That is,
In the feature amount memory 13, the size of the form using the identification code as an index and the feature amount for each character string (for each coordinate information) are written. Thereafter, the feature extraction unit 53 returns control to the common unit 51 once.

When the mode is read out from the mode area of the control memory 14, the common unit 51 sets the form registration mode to the blank mode.
Then, control is passed to step 5 (step A07).

In the form registration mode, the common feature extracting unit 55 which has taken over the control of the blank mode is in the blank mode, and therefore, the form size written in the feature memory 13 and all of the forms written in the feature memory 13 The feature amount of each character string coordinate information is registered in the dictionary memory using the form identification code as an index (steps A06, A07, A0).
9). In this case, the common feature extraction unit 55
A screen prompting the operator to input the form identification code or leave the form as it is (the form size written in the feature amount memory 13 + the form identification code as the form identification code) as the form identification code at the time of registration into the interface unit 17. Is displayed on the display device 4 via the. That is, in the dictionary memory, the form size and the characteristic amount for each coordinate information of the character string are registered using the form identification code as an index.

Next, a description will be given of the case of the filled-out mode in the form registration mode. When the operator uses the keyboard 3 to select and execute "the filled-out mode in the form registration mode" from the menu screen of the display unit 4. The common unit 51 of the control unit 11 registers the “completed mode in the form registration mode” in the mode area of the control memory 14. After the registration, the common unit 51 waits for input of a form.

When the operator inputs a blank form from the image scanner 2, the image scanner 2 optically scans the surface of the form and performs photoelectric conversion to generate a digital image of the form. The generated digital image of the form is read by the image input unit 52 of the common unit 51, and after correcting the form inclination and removing margins around the form, the digital image is stored in the image memory 12 (step A01 in FIG. 3). After being stored in the image memory 12, the image input unit 52 temporarily returns control to the common unit 51 in order to pass control to the feature extraction unit 53.

The feature extraction unit 53, which has taken over the control from the control unit 51, randomly generates an identification code not registered in the dictionary memory in order to identify the read form.
The form size detection unit 531 measures the size of the form held in the image memory 12, and writes the calculated identification code and the measured form size to the feature memory 13 (step A02). Further, the feature extraction unit 53 causes the character string feature detection unit 532 to detect the character pattern of the form held in the image memory 12 in units of character strings, and calculates a feature amount from the detected character pattern in units of character strings. (Steps A03 and A04). In this case, the feature extraction unit 5
The character string feature detection unit 532 for each character string for which the feature amount is calculated indicates coordinate information [eg, upper left corner coordinates (X, Y coordinates from the upper left corner) and horizontal length and height: character string Indicating the position of the area of the area] is calculated. The feature extraction unit 53 writes the calculated coordinate information and the calculated feature amount into the feature amount memory 13 using the previously calculated identification code as an index (step A05). In this way, the feature extracting unit 53 proceeds to step A03 until there is no more character string.
To A05 are repeated (step A06). That is, in the feature memory 13, the size of the form using the identification code as an index and the feature for each character string (each coordinate information) are written. Thereafter, the feature extraction unit 53
Returns control to the common unit 51 once.

When the common section 51 reads the mode from the mode area of the control memory 14, the common mode is set to the "completed mode in the form registration mode". A screen is displayed that prompts the operator to select whether or not there is a form to be filled in (a form to be added). If the operator selects “additional item”, the operator prepares a form having the same format as that of the previously processed form but with different contents, and reads the form from the image scanner 2. The control unit 11 repeats the operations of steps A01 to A07 until there is no more to be added (step A0
8). In this case, in step A05, the size of the form written to the feature memory 13 and the feature of each character string for each coordinate information are written to the area of the feature memory 13 assigned to the unused identification code.

In step A 08, if there is nothing to be added, common unit 51 passes control to common feature extraction unit 55. Since the common feature extraction unit 55 that has taken over the control in the filled mode is in the filled mode, the coordinate information of the character string in one identification code written in the feature amount memory 13 is in another identification code. All the character strings are compared to see if they match. If there is a match,
The common feature extraction unit 55 calculates the degree of similarity with the feature amount corresponding to the matched coordinate information, and if the similarity is equal to or greater than a predetermined value, it is determined that they match, and the coordinate information of the character string at that time and the feature amount Are all extracted (step A09). In this case, the common feature extraction unit 55 that calculates the similarity compares the pattern values for each window area, and divides the number of matches by the total number (the total number of window areas) to obtain a ratio obtained as the similarity. Ask. That is, if the calculated similarity is equal to or greater than a predetermined value (for example, 0.95), the common extraction unit 55 determines that they match. Therefore, even small-sized characters, which are difficult to decompose in character units, are compared using the pattern obtained as a character string, so that it is not necessary to decompose them in character units, processing is simplified, and form identification accuracy can be improved.

When the extraction is completed, the common feature extraction unit 55
Using the form identification code as an index, the feature amount of each extracted character string for each piece of coordinate information and the size of the form written in the feature memory 13 are registered in the dictionary memory 15 (steps A06, A07, A09). In this case, the common feature extraction unit 55 allows the operator to input the form identification code as the form identification code at the time of registration in the dictionary memory 15 or leave the identification code written in the feature memory 13 as the form identification code. A screen prompting the user to do so is displayed on the display device 4 via the interface unit 17.
That is, in the dictionary memory, the form size and the characteristic amount for each coordinate information of the character string are registered using the form identification code as an index.

Next, the case of the form identification mode will be described. When the operator selects and executes the "form identification mode" from the menu screen of the display unit 4 using the keyboard 3, the common unit 51 of the control unit 11 Registers "form identification mode" in the mode area of the control memory 14. After the registration, the common unit 51 waits for input of a form.

When the operator inputs a form from the image scanner 2, the image scanner 2 optically scans the surface of the form and performs photoelectric conversion to generate a digital image of the form. The generated digital image of the form is read by the image input unit 52 of the common unit 51, the inclination of the form is corrected, the margin around the form is removed, and the form is stored in the image memory 12 (step B01 in FIG. 4). After being stored in the image memory 12, the image input unit 52 temporarily returns control to the common unit 51 in order to pass control to the feature extraction unit 53.

The feature extracting unit 53, which has taken over the control from the control unit 51, generates an unused identification code in the dictionary memory to identify the read form.
31 is made to measure the size of the form held in the image memory 12, and the calculated identification code and the measured form size are written in the feature amount memory 13 (step B02).
Further, the feature extraction unit 53 causes the character string feature detection unit 532 to detect the character pattern of the form held in the image memory 12 in units of character strings, and calculates a feature amount from the detected character pattern in units of character strings. (Steps B03 and B0)
4). In this case, the character string feature detection unit 5 of the feature extraction unit 53
Reference numeral 32 denotes coordinate information for each character string for which the feature amount has been calculated [for example, the upper left corner coordinates (X, Y coordinates from the upper left corner) and the length and height in the horizontal direction: indicating the position of the area of the character string] Is calculated. The feature extracting unit 53 writes the calculated coordinate information and the calculated feature amount into the feature amount memory 13 as an index using the identification code generated earlier (step B0).
5). In this way, the feature extracting unit 53 repeats steps B03 to B05 until there is no more character string (step B06). That is, the feature amount memory 13 is written with the size of the form using the identification code as an index and the feature amount for each character string (each coordinate information). Thereafter, the feature extraction unit 53 returns control to the common unit 51 once. The above is the same as the processing in steps A01 to A06 in the operation in the form registration mode.

When the common section 51 reads the mode from the mode area of the control memory 14, it passes the control to the collation section 54 because the mode is set to the “form identification mode”. Collation unit 5
Reference numeral 4 sequentially reads out the feature amount written in the feature amount memory 13 and the feature amount of the same form size registered in the dictionary memory 15 and performs collation. For example, in the form C registered in the dictionary memory 15, first, after confirming the matching of the sizes, the coordinate values of the character strings of the form C are sequentially read out, and the coordinates of all the character strings detected from the processing target form are read. Compare with value. When the coordinate values match, the collation unit 54 calculates the similarity between the characteristic amounts of the character strings, and determines that the character strings match if the similarity is equal to or greater than a predetermined threshold value.
Further, when all the character strings of the form C registered in the dictionary memory 15 are all present on the form to be processed, the matching unit 54 determines the type of the form to be processed as the form C (step B07). . In this case, when the type of the form to be processed is determined, the matching unit 54 does not perform the matching with the unmatched form registered in the dictionary memory. Therefore, it is assumed that the forms with the highest appearance frequency are registered in the dictionary memory in order. The procedure in which the matching unit 54 calculates the similarity and regards the character string as a match is the same as the procedure in which the common feature extraction unit 55 calculates the similarity and regards the character string as a match, and a description thereof will be omitted.

Next, a description will be given using a specific example of the first embodiment of the present invention. An operation of registering a blank form A in a dictionary in a specific example as shown in FIG. 5 will be described. In this case, in the blank form A shown in FIG. 5, the character strings 2 and 3 have a small character size (for example, about 2 mm in height) and a small character spacing (for example, 0.5 m).
m) Therefore, this is an example of a character string that is difficult to cut out in character units from an image with a resolution of 200 DPI normally used in OCR.

First, a digital image is generated by reading a form by the image scanner 2, converted into a form image whose inclination has been corrected by the image input unit 52 and whose margin has been removed, and is stored in the image memory 12. In the feature extraction unit 53,
A form size and three character strings of character strings 1, 2, and 3 are detected from the digital image in the image memory 12. A feature amount is extracted from each of the three character strings, and is written to the feature memory 13. Further, the feature amount corresponding to the character string (coordinate information) is written from the feature memory 13 to the dictionary memory 06 using the form identification code given by the operator as an index.

Next, the operation of registering the filled-out form A in the dictionary in a specific example as shown in FIG. 6 will be described. First, the form A1 in FIG. 6 is read from the image scanner 2 via the image input unit 52, and is held in the image memory 12. The feature extraction unit 53 detects a form size and five character strings of character strings 1, character strings 2, character strings 3, character strings 4, and character strings 5 from the digital image in the image memory 12.
A feature amount is extracted from each of the five character strings and written into the feature amount memory 13. Next, the form A2 in FIG. 6 is similarly read from the image scanner 2 via the image input unit 52, is held in the image memory 12, and is recorded by the feature extraction unit 53 on the form size and the character strings 1 'and 2'. , Character string 3 ′, character string 4 ′, and character string 5 ′ are detected and written to the feature memory 13. Next, in the common feature extraction unit 55, the character strings included in the form A1 and the form A2 are compared. As a result, both the character string 1 and the character string 1 ′ have the same position information and feature amount. The same applies to character strings 2 and 2 'and character strings 3 and 3'. However, although the character string 4 and the character string 4 ′ have the same position information but different character patterns, the matching results of the feature amounts do not match. Similarly, the position information of the character string 5 matches the position information of the character string 5 ′, but the matching result of the feature amount does not match. Therefore, the feature amount for identification of the form A is character string 1 (character string 1 ′), character string 2 (character string 2 ′),
Only the character string 3 (character string 3 ′) is written to the dictionary memory 15.

In other words, by preparing a plurality of filled-in forms in the same format, the operator does not need to specify a character string to be registered in the dictionary or the like. It is possible to register a feature amount for form identification in a dictionary without increasing the burden.

When performing form identification using the dictionary memory 15 registered using the form shown in FIG. 5 or 6, the form to be processed is first read from the image scanner 2 via the image input unit 52. After the inclination of the form is corrected and the margin around the form is removed, the image is held in the image memory 12.
The feature extraction unit 53 detects the form size and all the included character strings from the digital image in the image memory 12. Then, after the form A registered in the dictionary is compared with the form size, if the form size matches, the position information and the character string of the character strings 1, 2, and 3 of the form A registered in the dictionary The feature amounts of the patterns are sequentially collated with the feature amounts of the character strings detected from the form to be processed. As a result, if all of the character strings 1, 2, and 3 match the character string detected from the form to be processed, the form to be processed is determined to be form A.

Next, a second embodiment of the present invention will be described with reference to the drawings. Referring to FIG. 8, a second embodiment of the present invention includes a control device 1 which is an information processing device such as a personal computer, an image scanner 2 for reading out a form,
A keyboard 3 for inputting characters and the like to the control device 1 and a display device 4 (for example, CR
T, LCD, etc.) and FDD which is a floppy disk drive
5 and a recording medium 6 on which a form identification program is recorded. The recording medium 6 may be a magnetic disk, a semiconductor memory, or another recording medium. FDD in this case
Reference numeral 5 denotes a storage control device corresponding to the recording medium.

That is, in the second embodiment, the FDD5
FIG. 2 is different from FIG. 2 in the first embodiment in that a recording medium 6 and a recording medium 6 are added. Therefore, the internal configuration of the control device 1 is as follows.
The description is omitted because it is the same as the inside of the control device 1 of FIG.

Next, the operation of the second embodiment of the present invention will be described. When the recording medium 6 on which the form identification program is recorded is inserted into the FDD 5 and the FDD readout program (for example, Explorer in the case of a personal computer) built in the control device 1 is executed, the form identification program on the recording medium 6 Is read into the program memory 16. That is, the form identification program is stored in the program memory 16, and the subsequent operation is the same as the operation in the first embodiment, and the description is omitted.

As described above, according to the present invention, when a character on a form is detected and converted into a feature value, the character is cut out in order to detect the character string unit without decomposing the character one by one. The work load of the operator who corrects the error can be reduced.

Further, according to the present invention, even when a form used for dictionary registration is a completed form or when only a part of the form layout is defined as in a postal transfer form, the feature memory is retained. Since the feature amount common to each set is extracted from the feature amounts of a plurality of sets of character strings and written into the dictionary memory, the operator's confirmation work can be reduced.

In the present invention, since character string features are collated on a character string basis without decomposing the character features on a character basis, small-sized characters for which pattern decomposition on a character basis is difficult are also used as identification features. And the form identification accuracy is improved.

[0063]

As described above, the first effect of the present invention is that a character pattern can be registered in a dictionary as a form identification feature even if the character pattern interval is narrow or a small character is used.
This is to improve recognition accuracy. The reason is that character pattern matching is performed not on a character basis but on a character string basis,
This is because there is no need to separate and detect character patterns in a character string. A second effect of the present invention is that the work load on the operator can be reduced. The reason for this is that, as described in the first effect, it is possible to register a character pattern with a narrow interval or a small character pattern that was impossible in the past, and there is no character string that cannot be registered. This is because a means for inputting an image and extracting a character pattern in a common character string is provided, so that only preprinting can be automatically selected and registered in a dictionary. This eliminates the need for the operator to select and specify a character string to be registered as a form identification feature.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram showing a flow of an operation at a functional block level when a control unit of FIG. 2 executes a form identification program stored in a program memory.

FIG. 2 is a block diagram illustrating a configuration of a first exemplary embodiment of the present invention.

FIG. 3 is a flowchart showing an operation of the control unit in FIG. 2 in a form registration mode.

FIG. 4 is a flowchart showing an operation of the control unit in FIG. 2 in a form identification mode.

FIG. 5 is a schematic diagram showing an example of a blank form;

FIG. 6 is a schematic view showing an example of a completed form.

FIG. 7 is a schematic diagram showing an outline of a feature extraction unit of FIG. 2 extracting a feature from a character string of a form.

FIG. 8 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.

9 is a schematic diagram showing an outline in which the image input unit in FIG. 2 corrects the inclination of the form and removes a margin around the form.

FIG. 10 is an enlarged view of the window area of FIG. 7 at the position of “house”.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 control device 2 image scanner 3 keyboard 4 display device 5 FDD 6 recording medium 11 control unit 12 image memory 13 feature memory 14 control memory 15 dictionary memory 16 program memory 17 interface unit 51 common unit 52 image input unit 53 feature extraction unit 54 Matching unit 55 Common feature extraction unit 531 Form size detection unit 532 Character string feature detection unit

Claims (10)

[Claims] 1. A form identification method for causing an information processing apparatus to identify a form, comprising detecting a character string from a digital image of an input blank form, extracting coordinate information for each detected character string, Furthermore, a feature amount is extracted while moving a window area for limiting a rectangular feature extraction range having the same height as the character string along the character string, and the coordinate information of the extracted character string and A form identification method characterized in that a form identification format is created by registering characteristic amounts in a memory. 2. A form identification method for causing an information processing apparatus to identify a form, wherein a character string is detected from each of digital images of a plurality of completed forms in the same format and having different contents, and the detected character is detected. Extract coordinate information for each column, and further extract a feature amount while moving a window area that limits a rectangular feature extraction range having the same height as the character string along the character string, The coordinate information and the feature amount of each of the extracted character strings are compared and collated for all of a plurality of filled-out forms having different entry contents in the same input format, and the coordinate information and the feature amount are compared. A form identification method characterized in that a form identification format is created by registering coordinate information and a feature amount of a matching character string in a memory. 3. A form identification method for causing an information processing apparatus to identify a form, wherein a character string is detected from a digital image of an input form, and coordinate information and a feature amount of each detected character string are extracted. A form identification method characterized by comparing the coordinate information and the characteristic amount of the extracted character string with the coordinate information and the characteristic amount of the character string registered in advance for each type of the form to identify the form type. . 4. A form identification method for causing an information processing apparatus to identify a form, wherein when an unfilled form is entered in a new format, a digital image of the entered form is stored in a first memory. Holding step;
A size measuring step of measuring a form size from the digital image held in the memory; detecting a character string from the digital image held in the first memory; calculating coordinate information of the detected character string; A feature amount calculating step of calculating a feature amount while moving a window region for limiting a rectangular feature extraction range having the same height as the character string along the character string on the character string; and the calculated coordinate information. A second storing step of storing the characteristic amount and the characteristic amount in a second memory; a first repeating step of repeating the characteristic amount calculating step and the second storing step until there is no longer a character string; When the first repetition step is completed, the form size held in the first memory and the feature amounts for each coordinate information of all the character strings held in the second memory are stored in the third memory. Form identification method characterized by comprising a registration step of recording. 5. A form identification method for causing an information processing apparatus to identify a form, wherein when a form written in a format is input, a first image storing a digital image of the input form in a first memory. A holding step; a generating step of generating an identification code for identifying the input form; a size measuring step of measuring a form size from a digital image held in the first memory; A character string is detected from the digital image held in the memory, the coordinate information of the detected character string is calculated, and a window area for limiting a rectangular feature extraction range having the same height as the character string is defined as a character. A feature value calculating step of calculating a feature value while moving along a character string on a column, and storing the calculated coordinate information and the feature value in a second memory based on the identification code. A second holding step, a first repeating step of repeating the feature amount calculating step and the second holding step until there is no longer a character string, and a form input after the first repeating step. When a form filled in in the same format is entered,
A second repeating step of repeating the generating step, the size measuring step, the feature amount calculating step, the second holding step, and the first repeating step;
After the end of the second repetition step, the coordinate information of the character string in the certain identification code stored in the second memory matches the coordinate information of the character string in another identification code stored in the second memory. And if there is a match in the comparison step, calculate the similarity with the feature amount corresponding to the coordinate information and match if it is equal to or greater than a predetermined value. An extraction step of extracting all the coordinate information and the characteristic amount of the character string, and extracting the characteristic amount for each coordinate information of all the extracted character strings and the form size written in the second memory after the extraction. And a registration step of registering the form in a memory. 6. A form identification method for causing an information processing apparatus to identify a form, wherein, when a form written in a format is input, a first image storing a digital image of the input form in a first memory. A holding step; a size measuring step of measuring a form size from the digital image held in the first memory; and a character string detected from the digital image held in the first memory, and A feature amount calculating step of calculating coordinate information and further calculating a feature amount while moving a window region for limiting a rectangular feature extraction range having the same height as the character string along the character string on the character string; And a second holding step of holding the calculated coordinate information and the feature amount in a second memory;
A first repetition step of repeating the feature amount calculation step and the second holding step until there is no longer a character string, and the second repetition step after the first repetition step is completed.
Of the character string held in the memory of the third form and the third form having the same form size as the form size stored in the second memory.
A comparing step of comparing the coordinate information of the character string of the form registered in the memory of the form, and calculating a similarity between the characteristic amounts of the character strings when the character strings match in the comparing step, and calculating the similarity between the character strings with a predetermined value or more. And determining the type of the form by determining that the character strings match, if any. 7. A form identification program for identifying a form to be processed by a computer, wherein when an unfilled form is input in a new format, a digital image of the input form is held in a first memory. A first holding step; a size measuring step of measuring a form size from the digital image held in the first memory;
Detecting a character string from the digital image held in the first memory, calculating coordinate information of the detected character string,
A feature amount calculating step of calculating a feature amount while moving a window region for limiting a rectangular feature extraction range having the same height as the character string along the character string, and the calculated coordinates; A second holding step of holding information and a feature amount in a second memory, a first repetition step of repeating the feature amount calculating step and the second holding step until there is no longer a character string; Upon completion of the first repetition step, registration for registering the form size held in the first memory and the feature amount for each coordinate information of all the character strings held in the second memory in a third memory A form identification program characterized by causing the computer to execute a process including the steps of: 8. A form identification program for identifying a form to be processed by a computer, wherein when a form written in a format is input, a digital image of the input form is stored in a first memory. 1, a generating step of generating an identification code for identifying the input form, a size measuring step of measuring a form size from a digital image stored in the first memory, A window area for detecting a character string from the digital image held in the first memory, calculating coordinate information of the detected character string, and further restricting a rectangular feature extraction range having the same height as the character string. Calculating a feature amount while moving the character along the character string on the character string; and storing the calculated coordinate information and the feature amount in the second memory. A second holding step of holding based on another code, and a first step of repeating the feature amount calculating step and the second holding step until there is no longer a character string
And when a form filled in in the same format as the entered form is input after the end of the first repeating step, the generating step, the size measuring step, the feature amount calculating step, and the 2, a second repetition step of repeating the holding step and the first repetition step, and after the end of the second repetition step, the coordinate information of the character string in the identification code stored in the second memory is stored in the second memory. A comparing step of comparing whether or not the character string matches the coordinate information of a character string in another identification code held in the second memory; and if there is a match in the comparing step, a feature corresponding to the matched coordinate information If the degree of similarity with the amount is equal to or greater than a predetermined value, it is determined that they match, and the extraction step for extracting all the coordinate information and the characteristic amount of the character string. And a registration step of registering, in the third memory, a feature amount for each coordinate information of all extracted character strings and a form size written in the second memory after the extraction is completed. A form identification program characterized by being executed by a user. 9. A form identification program for identifying a form to be processed by a computer, wherein when a form entered in a format is inputted, a digital image of the entered form is stored in a first memory. 1; a size measuring step of measuring a form size from the digital image held in the first memory; and a character string detected from the digital image held in the first memory by detecting a character string. A feature amount that calculates coordinate information of a column and further calculates a feature amount by moving a window region that limits a rectangular feature extraction range having the same height as the character string along the character string. A calculating step, a second holding step of holding the calculated coordinate information and the feature amount in a second memory, and a feature amount calculating step until there is no longer a character string. A first repetition step of repeating the first and second holding steps, a coordinate of a character string held in the second memory after the first repetition step is completed, and a character string stored in the second memory. Comparing the coordinate information of the character string of the form registered in the third memory with the same form size as the form size of the form, and determining the similarity between the characteristic amounts of the character strings when the character strings match in the comparison step. And determining the character string matches if the calculated value is equal to or greater than a predetermined value and causing the computer to execute processing including an identification step of identifying the type of the form. 10. A form identification method for identifying a form comprising an input device for converting a form into a digital image when the form is input, and a control device for performing a form identification process, wherein the control device reads the form from the input device. Image input means for storing a digital image of the form, and detecting the size of the form from the digital image of the form held by the image input means, and detecting the character pattern in units of character strings and converting them into feature amounts in units of character strings Feature extraction means for holding
When registering the format of the form to be input, a common feature extracting unit that extracts a feature amount common to each set from the feature amounts of a plurality of sets of character strings output from the feature extracting unit and registers the feature amount in a dictionary memory; When the format of the input form is identified, the characteristic amount of the character string of the processing target form output from the characteristic extracting unit and the characteristic amount of the character string held for each form in the dictionary memory in advance are used. A form identification method, comprising: a collation means for collating.